Background. Production AI systems require integrity, persistence, and verifiability beyond the statistical guarantees of foundation models. Existing orchestration frameworks (LangChain, AutoGen, CrewAI) provide composition without structural integrity, and existing memory systems (vector stores, knowledge graphs) provide retrieval without cryptographic guarantees. Methods. We specify the Chyren Sovereign Intelligence architecture, a Rust-first stateful AI orchestrator comprising approximately 33 crates in the Medulla workspace. Core components include: chyren-conductor (pipeline orchestration: Alignment -> AEON -> Provider -> ADCCL -> Ledger), chyren-aegis (policy enforcement via Merkle-signed manifests), chyren-adccl (drift detection at theta=0. 7), chyren-myelin (Qdrant-backed semantic memory), chyren-phylactery (cryptographically signed identity kernel with approximately 58k entries), and chyren-spokes (provider adapters with witness envelope signing). The Master Ledger uses Neon PostgreSQL with append-only signed entries. Results. We document the two-tier sovereign provider router (local Ollama for sensitive tasks; OpenRouter cascade for general workloads), the witness envelope signing scheme for cross-provider integrity (HMAC-SHA256 against YETTRAGRAMMATONSECRET), and the agent mesh layer for MQTT-based distributed orchestration. The architecture supports @secure pragmas routing final responses through a TEE driver with measurement attestation, and @authorialₚroxy pragmas enabling authorized ghostwriting with HUMANATTRIBUTIONREQUIRED ledger flags. Implications. The Chyren architecture provides a reference implementation of integrity-first AI orchestration, with formal guarantees enforced at the type-system level via Rust and verified via Lean 4 proof artifacts. The design demonstrates that statistical safety (RLHF) and structural safety (ADCCL + Master Ledger) are complementary rather than competing approaches.
Ryan W. Yett (Fri,) studied this question.